Search

The authors propose a heterogeneous floating-point (FP) computing architecture to maximize energy efficiency by separately optimize exponent processing and mantissa processing. The proposed exponent-computing-in-memory (ECIM) architecture and mantissa-free-exponent-computing (MFEC) algorithm reduce the power consumption of both memory and FP MAC while resolving previous FP computing-in-memory processors' limitations. Also, a bfloat16 DNN training processor with proposed features and sparsity exploitation support is implemented and fabricated in 28 nm CMOS technology. It achieves 13.7 TFLOPS/W energy efficiency while supporting FP operations with CIM architecture.

Generative adversarial networks (GANs) consist of multiple deep neural networks cooperating and competing with each other. Due to their complex architectures and large feature map sizes, training GANs requires a huge amount of computations. Moreover, instance normalization (IN) layers in GANs dramatically increase the external memory access (EMA). However, retraining GANs with user-specific data is critical on mobile devices because the pre-trained model outputs distorted images under user-specific conditions. This article proposes a GAN training accelerator to enable energy-efficient domain-specific optimization of GAN with user’s local data. Selective layer retraining (SELRET) picks out layers that are effective in enhancing the quality of the retrained model. Without image quality degradation, the SELRET reduces the required computation by 69%. Moreover, reordering layers for instance normalization (ROLIN) is proposed to reduce the EMA of intermediate data. Through the implementation of the proposed architecture, which splits and reorders the IN layers, 38.7% and 32.2% of overall EMA reduction are achieved in the forward propagation (FP) stage and the error propagation (EP) stage, respectively. The proposed processor is fabricated in a 65-nm CMOS process, showing 0.38-TFLOPS/W energy efficiency. The chip can retrain a face modification GAN with a custom dataset of 256 $\times $ 256 images over 100 epochs under 30 s while only consuming 274 mW. Compared to the previous FPGA implementation, this work improved the retraining performance and energy efficiency by 2 $\times $ and 39 $\times $ , respectively. As a result, the proposed accelerator enables GAN’s domain-specific optimization on a mobile platform.

A low power real-time visual object tracking (VOT) processor using the siamese network (SiamNet) is proposed for mobile devices. Two key features enable a real-time VOT with low power consumption on mobile devices. First, correlation-based spatial early stopping (CSES) is proposed to reduce the computational workload. CSES reduces ~56.8% of the overall computation of the SiamNet by gradually eliminating the background. Second, the dual mode reuse core (DMRC) is proposed for supporting both the convolution layer and the cross-correlation layer with high core utilization. Finally, the proposed VOT processor is implemented in 28 nm CMOS technology and occupies 0.42 mm2. The proposed processor achieves 0.587 for the success rate and 0.778 for the precision in the OTB-100 dataset with SiamRPN++-AlexNet. Compared to previous VOT processors, the proposed processor shows state-of-the-art performance while showing lower power consumption. The proposed processor achieves 64.1 mW peak power and 58.2 mW tracking power consumption at 32.1 frame-per-second (fps) real-time VOT on mobile devices.

Recently, Computing-In-Memory (CIM) processors have been proposed to achieve high energy-efficiency by reducing data movement and solving memory bottlenecks. Furthermore, a network with high accurate image classification has been introduced through the Absolute-Difference-Accumulation (ADA) operation instead of the multiplication-and-accumulation operation, which is widely used in DNN. ADA operation provides not only opportunities for high energy-efficient DNN accelerating by reducing multiplication but also a chance to reuse computation results. However, the previous CIM processor cannot reuse previous computation results for other computations. In this brief, we propose a high accurate and high energy-efficient ADA-CIM processor that with two key features: 1) computation reuse for low-power, resulting in a 49.5% CIM operation power reduction, and 2) low-cost sign prediction core with 3-bit activation and weight quantization for high utilization. From the two key features, the proposed ADA-CIM processor is simulated in 28 nm CMOS technology and occupies 3.78 mm2. It consumes 2.77mW and achieves 43.1 TOPS/W energy-efficiency with a high-accuracy of 91.62% at CIFAR-10 (ResNet-20).

Electric vehicles (EVs) can be considered as a revolution in the combustion industry with significant improvement in fuel utilization and decrease in pollution compared to combustion engines. However, by decreasing the size of the battery to reduce the cost, the frequency of charging EVs in a day increases. Therefore, to reduce the downtime required for charging EVs, wireless charging on the move can be an effective solution. In such a situation, paying for wireless charging on the move is an important issue. However, it can endanger the location privacy of users, since the EVs need to charge frequently in a day. In this paper, we first explain different methods of payment and problems with such payment methods in the case of wireless charging on the move. Then, we propose an efficient payment method based on `tokens' for wireless charging on the move, which minimizes the communications between service providers and users during the charging process. The proposed scheme prevents users and service providers from cheating, and it is robust to support different values for the price. Finally, we compare it with other payment methods that have been proposed for plug-in electric vehicles.

Today's high-end vehicles are technologically more sophisticated and smarter than the `old machines with engine and wheels'. These vehicles are anticipated to provide consumers with safe, reliable, and infotainment-rich driving experience through intelligent transportation system, realized through a technology referred to as vehicular ad hoc network (VANET). VANET offers a plethora of applications to the consumers ranging from safety to infotainment. Among other messages defined by the dedicated short range communication standard, cooperative awareness messages (CAMs) are of paramount importance for the VANET applications. In VANET, vehicles share their frequent mobility information in the form of CAMs with neighbors for cooperative awareness, traffic view construction, maneuver, and cruise control. In this work we target the CAM-based traffic information dissemination. VANET applications construct short-range local and long-range extend traffic views with the information contained in scheduled CAMs (also known as beacons). To construct aforementioned traffic views, naive approach is multi-hop communication; however, historically multi-hop communication does not scale efficiently in dynamic VANET due to computational and communication overhead. Therefore another alternative approach is essential to address the problems of the multi-hop communication in highly ephemeral VANET and to achieve maximum degree of the information dissemination in VANET. To fill the gap, in this paper we propose a geocast-based abstract piggybacking mechanism for traffic information dissemination in VANET through single-hop CAMs. In the proposed scheme, vehicles compress their local traffic view constructed through single-hop CAMs. The compressed local traffic view is referred to as traffic data unit (TDU). Every vehicle broadcasts its own TDU along with a number of TDUs received from other vehicles in the next scheduled CAM, in order to construct a long-range extended traffic view, also known as compressed data word (CDW). In other words, CDW is encapsulated into normal CAM in order to share it with the neighbors in the vicinity. The receivers of CDW use the received TDUs to construct the desired length of the long-range extended traffic view. The proposed scheme leverages only single-hop beacons data to construct both local and extended traffic views. Our simulation results show that the extended traffic view up to 1 km can be constructed with only single-hop beacons at the expense of 10 % of extra information with the regular CAM. The results also show that the effect of the extra overhead is not phenomenal from overhead standpoint. The proposed virtual multi-hop communication paradigm with single-hop CAM will reduce the overall communication thereby saving the bandwidth for other important safety related communications.

A computational approach to depth estimations using a color over lapped integral imaging system is presented. The proposed imaging system acquires multiple color images simultaneously through a single lens with an array of multiple pinholes that are distributed around the optical axis. This paper proposes a computational model of the relationship between the real distance of an object and the disparity among different color images. The proposed model can serve as a computational basis of a single camera-based depth estimation.

This paper proposes a new broadcasting system for the service-compatible 3DTV in which the 3D service can coexist with the conventional digital TV broadcast. In the proposed system, the commercial 3DTV service can be implemented via the existing DTV channel without utilizing the dedicated 3DTV system. This 2D/3D system interworks with the conventional system and can switch to 2D or 3D service according to the broadcast programming and schedule. The system also provides a mechanism that can prevent the synchronization mismatch between left and right video streams and between the stream and the associated signaling in the 2D/3D transition periods. The picture quality measurements are carried out based on the ITU-R recommended test to check the level of quality of service provided by the proposed scheme. The conformity tests are also performed with the conventional channel and the receiver for the DTV system to confirm the feasibility of the proposed one for the commercial service.

An energy-efficient maximum power point tracking (MPPT) circuit with a fast-tracking time for use with 800-μW PV energy harvesters is presented in this paper. The proposed MPPT circuit uses a successive approximation register MPPT algorithm, which has a power down mode and a fast tracking time, to achieve low power consumption and energy savings. The prototype MPPT circuit, which consists of analog-based circuits, has been implemented and fabricated in a 0.35-μm BCDiMOS process. The MPPT core occupies an area of 3 mm2 and consumes 4.6 μW of power. The tracking time is reduced by 69.4% and the stored energy is increased by 31.4% as compared to the conventional hill climbing-based MPPT algorithm under indoor conditions.

This letter presents a novel contrast enhancement approach based on dominant brightness level analysis and adaptive intensity transformation for remote sensing images. The proposed algorithm computes brightness-adaptive intensity transfer functions using the low-frequency luminance component in the wavelet domain and transforms intensity values according to the transfer function. More specifically, we first perform discrete wavelet transform (DWT) on the input images and then decompose the LL subband into low-, middle-, and high-intensity layers using the log-average luminance. Intensity transfer functions are adaptively estimated by using the knee transfer function and the gamma adjustment function based on the dominant brightness level of each layer. After the intensity transformation, the resulting enhanced image is obtained by using the inverse DWT. Although various histogram equalization approaches have been proposed in the literature, they tend to degrade the overall image quality by exhibiting saturation artifacts in both low- and high-intensity regions. The proposed algorithm overcomes this problem using the adaptive intensity transfer function. The experimental results show that the proposed algorithm enhances the overall contrast and visibility of local details better than existing techniques. The proposed method can effectively enhance any low-contrast images acquired by a satellite camera and is also suitable for other various imaging devices such as consumer digital cameras, photorealistic 3-D reconstruction systems, and computational cameras.

Image restoration plays a major role in improving image quality as a preprocessing step in various imaging systems. While conventional image restoration techniques, such as the constrained least-squares (CLS) and the Wiener filters, often exhibit either noise amplification or over-smoothing problem. On the other hand, advanced image restoration techniques, such as the iterative regularization and the combined Fourier and wavelet domain thresholding filters, are not suitable for real-time applications because of high computational overhead. For overcoming these problems, we present a novel vaguelette-wavelet decomposition (VWD) approach to real-time, space-frequency-adaptive image restoration based on frequency-adaptive shrinkage and directional wavelet bases. The proposed algorithm effectively reduces noise by adaptively shrinking wavelet coefficients based on alpha map and entropy, and restores a degraded image using the vaguelette-wavelet transform. In generating vaguelettes, we use the estimated point-spread function (PSF) and new directional wavelet bases. The extended set of experimental results confirm that the proposed algorithm can restore sharp details without amplifying noise, and is suitable for commercial low-cost, high-quality imaging devices, such as digital and computational cameras, visual surveillance systems, and consumer's camcorders.

This paper presents a novel wavelet-domain color image enhancement using filtered directional bases and frequency-adaptive shrinkage. Most traditional noise reduction methods tend to over-suppress high-frequency details. For overcoming this problem we first decompose the input image into flat and edge regions, and remove noise using the alpha map computed from wavelet transform coefficients of LH, HL, and HH bands. After removing noise in the flat region, we further remove noise in edge regions by adaptively shrinking wavelet coefficients based on the entropy. Moreover, we present a new directional transform using wavelet basis and Gaussian low pass filters. The wavelet coefficients of edge regions are inverse transformed by using the filtered wavelet bases. Experimental results show the proposed algorithm can reduce noise without losing sharp details and is suitable for commercial low-cost imaging systems, such as digital cameras, CCTV, and surveillance system.

In this paper, we present a novel image enhancement approach using a color shift model-based multiple color-filter aperture (MCA) camera for digital multifocusing. The proposed image enhancement algorithm consists of three steps; (i) cluster-based region-of-interest (ROI) estimation, (ii) image registration using phase correlation matching (PCM) and fusion, and (iii) image enhancement using spatially adaptive noise smoothing based on the alpha map. The image acquired by the MCA configuration contains color misalignment, which provides additional depth information of objects at different distances. This color misalignment can also provide additional information for blur estimation. The proposed cluster-based image segmentation method can effectively classify ROIs according to the distance from the camera. The segmented regions are aligned by using PCM, and they are fused to generate an in-focused image. For further enhancement of the color-registered image, we use spatially adaptive noise smoothing based on the alpha map. Experimental results show the proposed image enhancement method can significantly enhance the visual quality of the MCA output image, and can be fully or partially incorporated into multifocusing or extended depth of field (EDoF) systems in the form of the finite impulse response (FIR) filter structure.

In this paper, we present a novel real-time image restoration approach using a truncated constrained least-squares (TCLS) filter and spatially adaptive noise smoothing (SANS) algorithm based on alpha map for the extended depth of field (EDoF) system in an image signal processing (ISP) chain. The proposed TCLS filter and the alpha map-based SANS algorithm can be implemented in the Bayer-domain by using a general finite impulse response (FIR) structure. The TCLS filter coefficients are priori determined according to the given point-spread-function (PSF) by optical lens simulation, and deconvolution is performed in the Bayer-domain instead of the RGB-domain. The SANS algorithm can successfully remove noise in flat regions without affecting the sharply restored details. Based on the extended set of experimental results the proposed algorithm is proved to be able to restore an accurately focused image in real-time, and is suitable for commercial low-cost, high-quality imaging devices such as a digital camera and a camcorder.

In an RFID search protocol, a reader uses a designated query instead of an unspecified query commonly used in RFID authentication protocols. Due to this fundamental difference, techniques used in RFID authentication protocols may not be suitable for RFID search protocols. Tan et al.'s protocol, however, is based on techniques used in previous works such as using random values. In this paper, we propose two RFID search protocols, one based on static ID and the other based on dynamic ID, both which does not require additional measures to satisfy security requirements of RFID protocols. We achieve this by using counters.

This paper presents the field trial results of the terrestrial digital multimedia broadcasting (T-DMB) service in Korea. The T-DMB system, based on the Eureka-147 digital audio broadcasting (DAB) system, provides streaming video, CD-quality audio and data services at speeds of up to 300 km/h using H.264 video codec along with additional channel coding and interleaving schemes. In order to provide seamless high-quality services on various radio channels, we measured the bit error rate (BER), electric field strength (EFS), and subjective video quality using specially-designed mobile test equipment. From the field tests we obtained the required effective EFS (EEFS) for error-free reception and also derived the minimum EFS from the link budget analysis. Furthermore, we present the optimal transmit and receive parameters for constructing the T-DMB networks and also developing T-DMB broadcasting equipment and mobile receivers

This paper presents a feature-based object tracking algorithm using optical flow under the non-prior training (NPT) active feature model (AFM) framework. The proposed tracking procedure can be divided into three steps: (i) localization of an object-of-interest, (ii) prediction and correction of the object's position by utilizing spatio-temporal information, and (iii) restoration of occlusion using NPT-AFM. The proposed algorithm can track both rigid and deformable objects, and is robust against the object's sudden motion because both a feature point and the corresponding motion direction are tracked at the same time. Tracking performance is not degraded even with complicated background because feature points inside an object are completely separated from background. Finally, the AFM enables stable tracking of occluded objects with maximum 60% occlusion. NPT-AFM, which is one of the major contributions of this paper, removes the off-line, preprocessing step for generating a priori training set. The training set used for model fitting can be updated at each frame to make more robust object's features under occluded situation. The proposed AFM can track deformable, partially occluded objects by using the greatly reduced number of feature points rather than taking entire shapes in the existing shape-based methods. The on-line updating of the training set and reducing the number of feature points can realize a real-time, robust tracking system. Experiments have been performed using several in-house video clips of a static camera including objects such as a robot moving on a floor and people walking both indoor and outdoor. In order to show the performance of the proposed tracking algorithm, some experiments have been performed under noisy and low-contrast environment. For more objective comparison, PETS 2001 and PETS 2002 datasets were also used.

In this paper, electrohydrodynamic jetting is investigated in order to print ultra-fine dots and lines in drop-on-demand (DOD) mode, using micro-electromechanical system-based printhead with a piezoelectric actuator. In such hybrid system, jetting ultra-fine droplets in DOD mode, without applying an extremely high-voltage pulse, is possible as the meniscus is first disturbed by a piezoelectric actuator and the droplet is ejected by the applied electric field. As the amplitude of the drive waveform of the piezoelectric actuator is varied, droplets with volumes of 3.4 to 46.8 pL are realized. As the amplitude of the electric field is increased, the ejected droplets lengthen and at 8?kV, thin elliptical dots are printed. Although changing the jetting frequency from 0.1 to 2.0?kHz resulted in volume reduction from 9.4 pL down to 2.9 pL, the DOD characteristic is well maintained throughout. Such hybrid jetting characteristics enable the generation of diverse patterns in the printed electronics area.

A unique hybrid jetting system based on electrohydrodynamic and piezoelectric forces has been designed to verify the control of the drop velocity and to obtain ultrafine droplets with a high jetting frequency. Piezoelectric nozzles have been fabricated using silicon on insulator wafers and Pyrex glass employing a MEMS process and an anodic bonding process. The plate-type electrode and moving stage were used for the printing process. The droplet ejection mechanisms from the nozzle using the hybrid jetting system were captured by a high-speed camera synchronized with a trigger signal. The deformation of the meniscus and the jetting delay time in regard to the high operational firing frequency were investigated. It was found that controlling the droplet velocity without a change in the droplet volume and obtaining a smaller dot (59 ?m in diameter) in hybrid printing mode compared with inkjet printing mode (151 ?m in diameter) were possible. These results show this system's promising applicability to the fabrication of micro patterning for a wide range of printed electronics applications.